I'm building production document intelligence systems at Manthhan Software, fine-tuning SLMs and BERT-family models with LoRA, QLoRA, and PEFT for classification and structured extraction.
I own a RAG-based compliance checking system using AWS Bedrock embeddings, vector search, and LLM reasoning that achieves 94% compliance detection accuracy across more than 100,000 enterprise document pages. I also design deduplicated, multi-stage document workflows orchestrated with Prefect and deployed through AWS SageMaker, MLflow, Docker, and CI/CD.
Previously, I improved document extraction accuracy from 73% to 94%, built multimodal computer-vision pipelines that reduced manual review work by 40%, and led award-winning AI projects in edge intruder detection and EV energy optimization. My work has been published through IEEE and Taylor & Francis.

